In collaboration with John Bresina, Rich Washington, and Abdel-Illah Mouaddib.
This project is concerned with the design of a reactive meta-level controller that can optimize the operation of autonomous planetary rovers. Such rovers operate under tight resource constraints such as power, storage capacity, and communication bandwidth. The time available to carry out experiments is limited as is the overall lifespan of the rover. The amount of power that is available (between recharges) is limited and must be carefully managed. Storage capacity to be used for raw data and processed data before transmission to a control center is also limited. Some of the resources are renewable: batteries could be recharged and storage space could be freed once the date is transmitted. Another source of complexity in autonomous rover control is the high level of uncertainty regarding the consumption of resources by the rover's activities. For example, there is uncertainty about the amount of power and time required to bring the rover to a certain location, and there is uncertainty about the amount of storage that will be needed for a sequence of compressed images.
The combination of scarce resources and a high level of uncertainty present a complex meta-level control problem. The question is how to decide quickly during execution time which tasks should be executed and how to revise these decisions based on the actual progress being made, the availability of resources, and the remaining workload. Our approach to this problem is based on mapping each primitive activity (such as navigation, taking pictures, conducting experiments, etc.) into a progressive processing task structure that specifies alternative ways to accomplish each aspect of the activity. We then develop run-time control policies to choose the best way to execute each activity.